DIFFERENTIAL ESTIMATION OF THE GLOBAL MOTION PARAMETERS ZOOM AND PAN

A differential motion estimation technique for television sequences is presented which measures the parameters of global motion in the image plane due to zoom and pan of the camera. The algorithm is based on a two-dimensional signal model and a three-parameter model for motion description. The first global parameter describes a change in focal length corresponding to the ratio of the focal lengths before and after zooming. In the case of panning, two additional parameters have to be considered. Compared to known techniques, estimation errors due to displacement measurements are avoided as the algorithm measures the global parameters directly from the image signal. In order to reduce disturbing influences caused by moving objects or ambiguous matches, several sets of parameters are calculated locally, which are evaluated globally in a histogram.